Trajectory Prediction and Conflict Detection for Unmanned Traffic Management: a Performance Comparison of Machine-Learning-Based Approaches

The growing interest in Unmanned Aerial Systems has led, especially in recent years, to the need for a precise structure of rules and procedures aimed to guarantee safe traffic management. As an important type of dynamic data-driven application system, Unmanned Aerial Systems are widely used for civilian, commercial, and military applications across the globe and require a high level of autonomy. Indeed, the absence of the pilot onboard leads to stricter requirements regarding the level of accuracy for the Unmanned Systems, with the need to provide autonomous services such as Conflict Detection and Resolution during the strategic and tactical phases, avoiding possible collisions. In order to explore efficient and robust solutions for autonomous Collision Avoidance, an increasing research effort has been devoted to Trajectory Prediction for non-cooperative UASs. In this work, an analysis of existent solutions for Trajectory Prediction has been evaluated, with particular emphasis on methods based on Machine Learning (ML). Besides, four different Machine Learning methods based on Regression (Linear Regression, Regression Tree, Gaussian Process Regression, and Support Vector Machine for Regression) have been implemented for the prediction of the flight times of a single drone over planned paths. The four regression methods are used as benchmark for a proposed Deep Learning based approach, and the results have been compared.